Is the AI Bubble About to Burst?

When Great Technology Becomes a Terrible Investment

A great technology can still be a catastrophic investment if you pay too much when you buy in. That distinction — between the value of a technology and the price of a stock — is the single most important concept for anyone trying to learn about investing, whether you're a beginner or a seasoned professional. And right now, the AI industry is forcing that lesson at scale.

Anthropicis reportedly seeking an IPO valuation of around $2 trillion. To put that in perspective, The Economist notes that figure is roughly equivalent to the 10 biggest tech IPOs of all time added together. Its revenue, as of August, was running at approximately $65 billion annualised — impressive growth, but the valuation implies a price-to-revenue multiple of around 31 times. That number matters more than almost any other figure in this story, and we'll explain exactly why below.

Meanwhile, Nvidia — the company actually generating consistent profits from AI — is, relative to its expected earnings, the cheapest it has been in over a decade, despite a share price up more than 1,600% over four years. So the company selling the shovels looks cheap, while the companies digging with them are chasing trillion-dollar valuations. That asymmetry deserves serious analysis.

The Math Behind a $2 Trillion Valuation — And Why It Doesn't Work

Investment bankers typically value companies using one of two methods: a discounted cash flow (DCF) model, which projects future earnings and discounts them back to today's dollars, or a comparable company analysis, which applies valuation multiples from similar listed businesses.

Anthropicis awkward for both. There are very few comparable listed AI companies, and forecasting cash flows for a business whose revenue grew more than tenfold in a single year is largely speculative. So let's do the arithmetic honestly.

Assume Anthropic has:

• Zero cost of goods sold
• Zero operating expenses
• Zero taxes
• Zero R&D spending
• Every dollar of revenue paid directly to shareholders, forever

Even under those impossibly generous assumptions — discounted at the current 10-year Treasury yield, which recently touched 5.23%, its highest since 2004 — the present value of all that revenue in perpetuity comes to roughly $1.27 trillion. That's about $730 billion short of a $2 trillion valuation.

This is a variation of the argument Scott McNealy, co-founder of Sun Microsystems, made in 2002 after the dotcom crash. At 10 times revenue, he told Business Week, an investor would need Sun to operate with zero costs, zero taxes, and zero R&D for a decade just to break even — and even then, the time value of money meant it would actually take around 17 years. Anthropic is priced at 31 times revenue. McNealy's logic applies with compounding force.

The conclusion is uncomfortable but clear: almost the entire $2 trillion valuation is a bet on growth, not on current business fundamentals. And the higher interest rates go, the more explosive that growth needs to be to justify the price.

TAM Inflation: How Addressable Markets Grow Faster Than Economies

When standard valuation methods fall short, Wall Street reaches for the Total Addressable Market, or TAM — the theoretical maximum revenue a company could earn if it captured an entire industry. The metric has been around for decades, but it gained notoriety in the late 1990s when analyst Henry Blodget used it to argue Amazon, then trading at $240 per share, was worth $400. He was right about the stock in the short term. He was later banned from the securities industry for life.

Here is how TAM estimates for the AI industry have evolved in just four months:

• May: SpaceX cited a $22.7 trillion enterprise software and AI market
• September: Anthropic's IPO filing was reported to reference a $30 trillion market
• October: Morgan Stanley — reportedly a likely underwriter on Anthropic's IPO — estimated generative AI could address $60 trillion, or roughly half of global annual economic output

For context, global GDP is approximately $105 trillion. The AI industry's self-reported addressable market grew by $37 trillion in four months — faster than Anthropic's revenue and considerably faster than the global economy it's supposed to be carved out of.

History suggests healthy scepticism is warranted. Uber touted a $122.3 trillion TAM at its 2019 IPO. Its annual revenue today is under $60 billion — meaning it has still captured less than 0.05% of that market. WeWork claimed a $3 trillion addressable market before filing for bankruptcy. The market, it seems, went unaddressed.

Anthropicis own internal research models an extreme scenario in which AI adds over $10 trillion to US GDP by 2030, which analysts have translated into approximately $100 trillion in equity value. Beyond that sits the concept of recursive self-improvement — AI that designs better AI, which designs even better AI, creating so much economic value that asset prices lose coherent meaning. That is, technically, one way to hit a price target. It is not, however, a basis for a DCF model.

A Circular Funding Structure That Demands Scrutiny

Perhaps the most structurally important part of the current AI boom is not the valuations themselves, but who is paying whom — and how circular those arrangements have become.

Consider SoftBank's data center subsidiary, SB Energy, which filed for a $50 billion IPO despite having never switched on a single data center. Its prospectus disclosed it would need $174 billion to build what it has already committed to building. It recently issued the largest junk bond offering on record, yielding 9.75%. Its lenders won't permit a dividend until 2029.

Now trace the money:

1. SoftBank borrows at junk rates (9.75%) to fund investments into OpenAI
2. OpenAI — which the FT reports expects to burn through nearly $280 billion by 2030 — uses that capital partly to lease SB Energy's Ohio campus for 20 years
3. That lease gives SB Energy the revenue it needs to justify its $50 billion valuation
4. Nvidia bought $1.5 billion of SB Energy shares in August at a 10% discount to the IPO price, agreed to buy another $1.5 billion at listing, and has guaranteed up to $105 billion for the Ohio campus — in exchange for that campus running exclusively on Nvidia hardware for 20 years
5. OpenAI also holds warrants in SB Energy that pay out if SB Energy's valuation reaches $80 billion

Separately, according to the Wall Street Journal, SpaceX is renting compute capacity to Anthropic for $1.25 billion per month. And according to analysis cited on the ProfG Markets podcast, 85% of Amazon's net income in its most recent results came from unrealised gains on its stakes in Anthropic and OpenAI. For Alphabet (Google), the figure was 87%, derived from stakes in SpaceX and Anthropic.

This is not a collection of independent businesses making money from customers. It is a network of large companies booking gains on each other's valuations. The moment external capital stops flowing in — because rates stay high, because IPO markets stay cold, or because institutional investors grow cautious — the internal logic of this structure faces a serious stress test.

Rising Rates: The Variable That Changes Everything

For anyone learning about investing for the first time, understanding the relationship between interest rates and asset valuations is foundational. When rates rise, the present value of future profits falls. A dollar of profit expected in 10 years is worth significantly less today when discounted at 5% than when discounted at 2%. For companies whose entire valuation rests on profits a decade or more away, this effect is dramatic.

The 10-year Treasury yield recently hit 5.23%, its highest level since 2004. Matt Kennedy of Renaissance Capital has described this as a "double whammy" for high-growth tech: it reduces the present value of distant profits, and it increases the cost of borrowing to build the infrastructure those profits depend on.

This helps explain a puzzling paradox: the NASDAQ is at record highs, US business output is growing at its fastest pace in five years, and yet IPO after IPO is being pulled or delayed. WholeTech, a nuclear power company serving AI data center demand, postponed its IPO citing market conditions. SB Energy filed before Labor Day and has still not begun marketing. Only three IPOs have priced since Labor Day, and five of the 10 largest listings this year are now trading below their offer prices.

The last time the Fed raised rates into a similarly strong economy was 1999. Investors who remember what followed tend to apply the current data with particular attention.

What Nvidia's Valuation Actually Tells Us

While Anthropic seeks $2 trillion on $65 billion in revenue, Nvidia presents a counterintuitive picture. Its share price is up more than 1,600% over four years, making it the world's most valuable company by market capitalisation. Yet Bloomberg notes that relative to the profits it is actually expected to generate, Nvidia is the cheapest it has been in over a decade.

This is what happens when earnings grow fast enough to outpace even a spectacular share price. Nvidia's data centre revenue alone exceeded $22 billion in a single quarter in its most recent results. It is generating real, auditable, taxable profit from companies paying real money for chips that are in use today.

For investors learning how to evaluate growth stocks, the Nvidia-versus-Anthropic comparison is instructive. Both are AI companies. Both are growing quickly. But one has a product with a clear price, a clear customer, and a clear profit margin. The other has a product with real users but a valuation that requires assumptions that make dotcom-era arithmetic look conservative.

That does not make Anthropic worthless, or even overpriced at every conceivable scenario. But it does mean the price you pay matters enormously — arguably more than the technology itself.

Key Takeaways for Investors

• Valuation multiples matter more than growth narratives. A 31x revenue multiple requires near-perfect execution over a long time horizon just to break even at current interest rates.
• TAM estimates are not revenue forecasts. Treat them as marketing, not analysis. Compare the claimed market size to actual GDP figures as a basic sanity check.
• Circular funding structures amplify risk. When major players are booking gains on each other's valuations rather than on external customer revenue, the system is more fragile than headline numbers suggest.
• Interest rates are the discount rate on the future. Higher rates hurt growth stocks disproportionately. At 5%+ on the 10-year, that effect is material for any company whose profits are 5-10 years away.
• Profitable companies with real earnings deserve different analysis than pre-profit platforms. Nvidia and Anthropic are not the same type of investment, even though both operate in AI.

Frequently Asked Questions

How do investment bankers value a company before its IPO?

Bankers typically use two methods: a discounted cash flow (DCF) model, which projects future profits and discounts them to today's value using an interest rate, and a comparable company analysis, which applies valuation multiples from similar listed businesses. For pre-profit companies in new industries, both methods require significant assumptions, which is why IPO valuations often rely heavily on growth narratives and total addressable market estimates rather than current fundamentals.

What is a TAM, and why should investors be cautious about it?

TAM stands for Total Addressable Market — the theoretical maximum revenue a company could generate if it captured an entire industry. It is a legitimate planning tool but a poor valuation basis. Companies rarely capture more than a small fraction of their stated TAM. Uber claimed a $122.3 trillion TAM at its 2019 IPO; its revenue today is under $60 billion. WeWork cited a $3 trillion market before going bankrupt. TAM figures in IPO filings should be treated as optimistic projections, not financial forecasts.

Why do rising interest rates hurt AI and tech stock valuations so much?

High-growth technology companies are valued largely on profits expected years into the future. The higher the interest rate used to discount those future profits back to today's dollars, the lower their present value. At a 5% discount rate, a dollar received in 10 years is worth about 61 cents today. At 2%, it's worth about 82 cents. For companies where most of the valuation is in distant future earnings, a 3-percentage-point rise in rates can reduce a fair valuation by 20-30% or more, even if the underlying business hasn't changed.

Is Nvidia's valuation a reliable benchmark for judging AI investment value?

Nvidia is a useful reference point precisely because it generates substantial, auditable profits from AI chip sales today — not in five years. Its price-to-earnings multiple is, relative to its history, at a decade low despite massive share price appreciation, because earnings have grown faster than the stock. This makes it structurally different from pre-revenue or pre-profit AI platforms. Comparing Nvidia's valuation to an Anthropic IPO price would be misleading; the risk profiles and cashflow structures are fundamentally different.

What does the dotcom era teach us about paying high revenue multiples?

Sun Microsystems co-founder Scott McNealy famously demonstrated in 2002 that at 10 times revenue, an investor would need a company to operate with zero costs, zero taxes, and zero R&D for roughly 17 years — even before accounting for the time value of money — just to recover their investment. Anthropic's reported IPO valuation implies approximately 31 times revenue. Historical precedent strongly suggests that buying into companies at high revenue multiples requires either exceptional subsequent growth or significant patience — and often results in long periods of negative returns even when the underlying technology succeeds.

This article is for informational purposes only and does not constitute financial advice. Always consult a qualified financial professional before making investment decisions.